Re: CONFIG_PREEMPT and server workloads

[Dipankar Sarma <dipankar@in.ibm.com>]

On Fri, Mar 19, 2004 at 11:30:16AM +0100, Takashi Iwai wrote:
> > The unavoidable worst case is in the RCU callbacks for dcache shrinkage -
> > I've seen 25 millisecond holdoffs on the above machine during filesystem
> > stresstests when RCU is freeing a huge number of dentries in softirq
> > context.
> 
> hmm, this wasn't also evaluated in my tests.
> it's worthy to try.  thanks for the info.

Please do. This would be a very interesting data point.

> 
> > This if Hard To Fix.  Dipankar spent quite some time looking into it and
> > had patches, but I lost track of where they're at.
> 
> couldn't this tasklet be replaced with workqueue or such?

Close. I am using per-cpu kernel threads to hand over the rcu callbacks
if there are too many of them. It depends on CONFIG_PREEMPT on better
latency from there onwards.

I can use a workqueue, but I needed the flexibility experiment with
the kernel thread for now.

Thanks
Dipankar